> > The human brain-body system (we need both for AGI) has a complexity of > about 5 x 10^9 bits, according to the best known compression > algorithms applied to the human genome. It is equivalent to about 300 > million lines of code.
But please note the distinction between acquiring knowledge by > learning (evolution, observation, induction, communication, storage in > books and computers), and manufacturing knowledge by modifying our > code. We can only manufacture computing power, by reproducing and > technology. In turn, this increases the rate at which we can learn. I do note the difference between acquiring and manufacturing information. I was pointing out the difference, in fact. Those 5x10^9 bits or 300 million lines of code in our genome were acquired by (multi-agent) learning, and encoded into the genome. What is to stop any learning agent from acquiring information and then modifying its own source to encode that information, thereby increasing its effectiveness at computation, learning, or problem solving within the domain it has acquired information about? It's not free information, and it's not manufactured. It is an increase to the agent's own complexity through acquisition of information from the environment, encoded into its own structure. Humans are an example of this process. That said, no, I don't buy into an agent isolated in a box that magically goes "foom". The agent must acquire information to enhance its own capabilities, and for that to happen there must be interaction with the environment, which limits the rate of acquisition of such information. There is no reason to assume, however, that the maximal rate of information acquisition is no higher than the approximate level at which humans operate. On Wed, Mar 5, 2014 at 5:22 PM, Matt Mahoney <[email protected]>wrote: > On Tue, Mar 4, 2014 at 7:22 PM, Aaron Hosford <[email protected]> wrote: > > We are intelligent, and we do not start with many bits of knowledge. > > The human brain-body system (we need both for AGI) has a complexity of > about 5 x 10^9 bits, according to the best known compression > algorithms applied to the human genome. It is equivalent to about 300 > million lines of code. > > > We are > > general purpose learners using the magic of intelligence to increase our > own > > intelligence, and we do it all the time. > > If you mean using computers to offload some of our thinking, and > taking credit for it, then I agree. It doesn't really matter if Google > makes people smarter or whether Google itself is smarter, as long as > the total intelligence is increasing. > > But please note the distinction between acquiring knowledge by > learning (evolution, observation, induction, communication, storage in > books and computers), and manufacturing knowledge by modifying our > code. We can only manufacture computing power, by reproducing and > technology. In turn, this increases the rate at which we can learn. > > I assume that you agree with me that when MIRI raises the scenario of > AI in a box that goes FOOM, we are really referring to the threat of > self replicating agents such as computer viruses, robots, > nanotechnology, or engineered organisms. The agents need not be > individually intelligent, just collectively when you sum over their > knowledge and computing power. > > Whatever these agents are, they are not general purpose reinforcement > learners with fixed goals. Rather, they are agents designed by people > to carry out specific functions. If they deviate from these functions, > you can blame design errors and evolution. > > -- > -- Matt Mahoney, [email protected] > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/23050605-2da819ff > Modify Your Subscription: > https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com > ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
